Uneven distribution of data on historical nodes

Apache Druid Q&A Google Groups
#1

Hello,

When loading data from S3 after hadoop indexing we often see an uneven distribution of data on our historical nodes. Recently one node was more than 90% full while others were ~60% full. Is this a problem? Does it suggest a configuration problem?

Thanks for your help,
Morri Feldman

image.png

#2

Forgot to mention that we are running imply-2.0.0 / Druid 0.9.2

#3

It could just mean that your cluster is balancing slowly. There’s a throttle called “maxSegmentsToMove” that you can edit if you click the pencil in the coordinator console. The default value is pretty low.

#4

Thanks @Gian

Unfortunately, we are already running with maxSegmentsToMove set to 300.

#5

Hi,

I have the same problem with you. Have you solved this problem?

Could you tell me how to solve this problem?

Thank you.

在 2017年5月19日星期五 UTC+8上午1:37:22，mo…@appsflyer.com写道：

#6

Hi Linjing,

We never really solved it. I just checked and two of our clusters have an even distribution of data on the historical nodes, but on one cluster the nodes vary by ~20% in how much data they are holding.

Best,

Morri